<![CDATA[This study compares the convergence and stability of two artificial neural network (ANN) training methods Pure DFNN and BackPropagation for classifying the severity of stator winding faults in induction motors. The evaluation was conducted through 10 trials for each model, with success criteria defined by achieving a Mean Squared Error (MSE) close to the threshold of 0.0001. The results indicate that Pure DFNN achieved the fastest convergence with 294 iterations and an MSE of 0.000101230. However, its iteration range was highly fluctuating, reaching up to 3514 iterations, indicating sensitivity to weight initialization. In contrast, the ANN BackPropagation method demonstrated more stable convergence, with all trials falling within a narrow range of 226 to 449 iterations and achieving the best MSE of 0.000100122. Both methods succeeded in producing minimal final error values. Nevertheless, the findings suggest that ANN BackPropagation is superior in terms of stability and robustness, while Pure DFNN may offer faster convergence in certain cases but lacks consistency. These results are significant for selecting appropriate ANN training algorithms in induction motor fault diagnosis systems that require long-term reliability.]]>
Copyrights © 2026
